My first 4 months of value betting
I have started to bet following a model this season. In this post I am going to describe my first
attempts, how I built a strategy and the results I reached.
After spending months developing a model for expected points I decided in September to spend a
little time and money to validate it in the real world. I deposited 200 £ and started to bet.
At first I tried with Overs but I quickly found out that was not the best strategy as I ended up
losing 10 £ in my first 9 bets. The reason is that Alfa Data model predicts expected points much
better than expected goals. So I started to focus on 1X2 betting, selecting higher odds and putting
a lower amount on each bet. I placed 3 bets in September following this strategy. Here they are.
Country |
Division |
HomeTeam |
AwayTeam |
Prediction |
Wager |
Odds |
Result |
Italy |
Serie A |
Udinese |
Sampdoria |
1 |
5 |
2.719 |
8.55 |
Germany |
Bundesliga |
Eintracht Frankfurt |
Stuttgart |
1 |
5 |
2.080 |
5.40 |
Spain |
La Liga |
Leganes |
Atletico Madrid |
X |
5 |
3.710 |
13.55 |
All successful. I bet 15 £ and won more than 25 £. That was the beginning of my strategy.
My attempts and mistakes in the month of September have thought me some lessons that I think are
valuable.
- Always bet something between 2-5% of your bankroll on a single bet. Adjust the amount monthly.
- Only bet on Leagues you know (or you have a model for).
- Never ever do accas.
- When in doubt bet on a draw.
These suggestions are valid indipendently from the model used. You can follow them even if you bet
using your guts.
Let me stress again how is important not to do accas. Accas are the best way to lose money. Betting
on a combination of outcomes increases greatly your chances to lose. Even if every bet has very low
odds the chance that one of them goes wrong increases exponentially with the number of bets in the
acca. It's simple maths. And the bookies know it.
After the first few bets, ath the end of September I had a little more than my initial bankroll of
200 GBP. I decided to place 5 GBP for every bet. This is indeed 2.5% of the bankroll.
The bets above are useful to show the best strategy for this model. First, I decided for a 1X2 type
bet. This is perfect for a model that gives you expected points. Second, I identify the over and
under performing teams and see if any of them are playing against each other. Third, I check if the
odds are high enough for this type of bets. Given past performance of the model I find it's
profitable to bet on thse type of matches only when the odds are above 2 more or less. Last, you
bet on the team that is underperforming and against the team that is overperforming. You are
basically betting on regression to the mean for both teams.
I bet the same amount on each bet. I adjusted the amount bet every month. I kept it at 5 GBP for the
whole of October, increased at 7 in November, decreased at 6 for the first bets in December and
again increased at 7 for the latest bets. In the future I plan to adjust the amount every week.
The performance of this strategy over the month of October, November and December is shown below.
Month |
Number of Bets |
Total Stake |
Won |
ROI |
October |
26 |
130 |
32.70 |
25% |
November |
30 |
182 |
-11.69 |
-6.4% |
December |
12 |
81 |
81.09 |
100% |
Total |
68 |
393 |
102.1 |
26% |
Monthly Average |
22.3 |
131 |
34.0 |
39.7% |
The sample includes 3 months betting for a total of 68 bets placed. Clearly not a very big sample
but big enough to draw some conclusions.
The model seems to have a predictive power. The number of correctly predicted bets is higher than
50%.
The model is profitable. The total profit made is over 100 GBP, which means I increased my bankroll
from 200 GBP up to 300 GBP in just 3 months. The profit per bet is 1.50 GBP per bet.
The model has high variance. Looking at the ROI month by month it's pretty clear that the value
fluctuates a lot and it will take another couple of months of bets to stabilise.
Overall these 4 months have been productive. The first month was used to tune the strategy and
identify the best way to extract value from the model. Luckily it ended up positive too.
In the table below you can see the overall results including September.
Month |
Number of Bets |
Total Stake |
Won |
ROI |
September |
12 |
75 |
18.43 |
24.5% |
October |
26 |
130 |
32.70 |
25% |
November |
30 |
182 |
-11.69 |
-6.4% |
December |
12 |
81 |
81.09 |
100% |
Total |
80 |
468 |
120.5 |
25.7% |
Monthly Average |
20 |
117 |
34.0 |
35.9% |
Looking back at these 4 months of experiemnts I think the results are positive. Not only financially
but also in terms of knowledge gained. I learnt that a model is not enough to design a betting
strategy. Bankroll management and being able to identify bets with value are even more important. I
see betting as an investment opportunity. It's risky but not more than investing in stocks. At
least for betting I have a model and some domain knowledge.
In the next months I will keep monitoring the metrics of the strategy and I can only hope that my
performance will be the same, that would mean I will have a bankroll of 750 £ by June. My
expectations are lower, based on past performance, the ROI of the model is somewhere closer to 5%
than the current 25%.